Hard check simulator for loan rules
Try the generated demo or drop in past applications with a column showing whether each rule fired and a column for the outcome. See which rules to set as KO, approve but investigate or leave out, and how KO changes the bad rate and applications rejected, using your chosen multiple of the base bad rate, such as 6% or more at a base bad rate of 2%.
Your applications
10,000 applications read, 317 bad, 8,178 good and 1,505 with no outcome.
The file is read in this browser and never uploaded, and only counts leave the page if you request the result by email.
In the outcome column, bad, default, fraud, 1 or yes count as bad, and good, 0 or no count as good. Anything else, such as an empty cell for a loan that was not booked or has not run long enough, counts as no outcome. Every other column containing only 1 and 0, yes and no, true and false, or blanks is read as a rule that fired or did not.
10 rule columns found: address_not_matching_bureau, applied_between_1_and_5_am, device_on_3_or_more_applications, device_shared_with_blacklisted_phone, email_younger_than_7_days, employer_phone_is_applicant_phone, id_document_expires_within_30_days, income_above_employer_p90, phone_on_blacklist, same_ip_as_rejected_application
The demo has 10,000 generated applications and ten rule flags, not a lender's data.
Bad rate after KO
The 3 KO rules reject 3.0% of applications and stop 23.3% of the bad ones. The bad rate goes from 3.7% to 3.0%.
The one rule sends 5.8% of applications to investigation after approval, and the bad rate among them is 7.4%.
Lift by rule
Each bar shows how many times the base bad rate the rule's bad rate reaches, and the dashed marks show the investigation and KO multiples.
Approval rate cost
The KO rules take the approval rate from 100.0% to 97.0%, a fall of 3.0 points.
Each bar shows the approval rate points a rule would take away as a KO, counting only applications the KO rules above have not rejected, and grey and orange bars show the cost if that rule were made a KO.
Rule results
| Rule | Hits | Hit rate | Bad rate | Lift | New hits | New hits' bad rate | Verdict |
|---|---|---|---|---|---|---|---|
| phone_on_blacklist | 64 | 0.6% | 63.2% | 16.9× | 64 | 63.2% | Set as KO |
| device_shared_with_blacklisted_phone | 66 | 0.7% | 49.2% | 13.2× | 22 | 4.5% | Covered by KO rules above |
| same_ip_as_rejected_application | 18 | 0.2% | 35.7% | 9.6× | 17 | 30.8% | Too few to tell |
| device_on_3_or_more_applications | 137 | 1.4% | 22.6% | 6.1× | 136 | 21.9% | Set as KO |
| employer_phone_is_applicant_phone | 102 | 1.0% | 16.9% | 4.5× | 100 | 14.9% | Set as KO |
| email_younger_than_7_days | 596 | 6.0% | 9.1% | 2.4× | 575 | 7.4% | Approve but investigate |
| applied_between_1_and_5_am | 393 | 3.9% | 6.4% | 1.7× | 380 | 5.0% | Covered by KO rules above |
| address_not_matching_bureau | 790 | 7.9% | 5.5% | 1.5× | 757 | 3.9% | Leave out |
| income_above_employer_p90 | 552 | 5.5% | 4.7% | 1.2× | 542 | 3.2% | Leave out |
| id_document_expires_within_30_days | 223 | 2.2% | 2.1% | 0.6× | 216 | 1.1% | Leave out |
Rules are judged from the highest lift down. A rule whose bad rate meets the KO multiple on its own is set as KO only if the bad rate of its new hits also meets it.
Test a new rule
Email your results
The same check can run on past lending decisions, comparing each rule's hits with how the loans turned out and testing changed KO rules on those decisions before they go live. Start free
See the formulas
Hit rate = Hits ÷ All applications Bad rate = Bad hits ÷ Hits with outcome Lift = Bad rate ÷ Base bad rate New hits = Hits, Not rejected by KO above Set as KO: Lift ≥ KO multiple, New hits' bad rate ÷ Base bad rate ≥ KO multiple Approve but investigate: Lift ≥ Investigation multiple, New hits' bad rate ÷ Base bad rate ≥ Investigation multiple Too few to tell: Hits with outcome < Minimum applications with outcome
Outcomes are known only for applications that were approved and ran long enough, so a rule that rejected an application in the past cannot be judged from this file. The results show what each rule picked out in these past applications, and do not predict what it will do next month.
Hard check rules and bad rate
A hard check is a rule that rejects an application outright. This tool reads past application outcomes and rule flags, then compares each rule's bad rate with the base bad rate.
The result separates rules that meet the KO threshold from rules to approve but investigate or leave out. It also shows how the KO rules change the bad rate and the share of applications rejected.
Common questions
What does a hard check simulator show?
A hard check is a rule that rejects an application outright. The simulator compares the rule's bad rate with the base bad rate, then checks whether its new hits meet the same threshold.
How do I choose KO rules for loan applications?
The tool sets a rule as KO when its lift and the lift of its new hits both reach the chosen KO multiple, with enough applications that have an outcome. You can set the multiple and the minimum count in the settings.
What bad rate should a hard check reject at?
The KO threshold is the chosen multiple of the base bad rate. For example, with a 2% base bad rate and a multiple of 3, a rule needs a bad rate of at least 6%.
How is lift calculated for a fraud rule?
Lift is the rule's bad rate divided by the base bad rate. It compares the rate among the rule's hits with the rate across the portfolio.
How many applications does a rule need to assess its bad rate?
The tool marks a rule too few to tell when fewer than the chosen minimum number of its hits have an outcome. You can change the minimum in settings.